An International Multilab Investigation on the Impact on Sexual Arousal of Believing that the Content is AI-Generated
Warning: This repository contains graphic stimuli (erotic images) from the open-access scientific database NAPS-ERO. This is for scientific purposes only. If you are minor, please quit this repository immediately.
This international and interdisciplinary project brings together neuroscientists, psychologists and philosophers to study the effect of reality beliefs on sexual arousal.
Imagine that Alex receives a love poem from an admirer. She/he/they loves the rhymes and feels flattered. However, they then find out it was not actually made by the admirer, but rather by ChatGPT. Would they still be flattered? Now, turn to a more prosaic situation: imagine that Alex met Cal via an online dating app. They feel in touch and begin to "sext" (i.e. to exchange intimate pictures). Alex likes Cal, and is aroused by the pictures. But then, one day, Cal confesses that the pictures are not photos, but rather sophisticated AI-generated images. Would Alex still feel equally aroused by the images after having learned they are not real?
Due to the rapid development of accessible software for generating AI content resembling real persons, this scenario seems increasingly realistic, making the need to address such a question quite compelling, as they might affect societal issues such as non-consensual deepfake pornography [10]. To tackle it, we need to unravel whether believing that someone is real (as opposed to fictional) changes our erotic arousal towards us. Several published papers attest that an equivalent content is more emotionally invested when believed to be real than when believed to be fake [1-6]. More recently, it has been reported that presenting a face as AI-generated dampens its perceived trustworthiness [7]. In a previous pre-registered pilot study [8], some of us found that pictures of people in underwear are judged on average slightly sexier when they are believed to be real photos rather than AI-generated (either due to subjects' own judgments or to an experimental manipulation). The present study aims at investigating the top-down influence of believing that a picture represents either an actually existing person or a fictional one on the sexual arousal it triggers in the onlooker. It aims at replicating the results of [8] via a multi-country collaboration, employing validated stimuli from the EroNAPS database [9]. We also hope to investigate whether the effect found with self-report data is aligned with objective physiological data, as obtained from skin conductance and cardiac activity measurements.
Participants will be shown 80 pictures from the EroNAPS. The stimuli include erotic photos of men and women, as well as non-erotic photos. While all stimuli are real photos, subjects will be told that half of them (a random set) are AI-generated content via a cue (written prompt) presented right before the stimulus itself. For each stimulus, subjects are asked how sexy they think it would be judged by an average onlooker and how arousing they find it from their own perspective. We aim to collect 100 participants per country (total = 400). We will collect some basic demographic information (e.g. gender, age, education, sexual orientation) and some inventory (e.g. Socio-Sexuality scale, General Attitude toward Artificial Intelligence scale). Participants will be asked to verify their age, so that underage subjects will not be exposed to erotic material. We predict that the cue produces a noticeable effect on arousal ratings, but not on sexiness ratings (or at least, not to the same extent). We will also investigate whether some features of the respondents (e.g. their gender) or of the stimuli (e.g. direct gaze, which might trigger attribution of mental state) mediate the ratings.